Collaborative process maturing support by mining activity streams. iknow 2015 Prof. Dr. René Peinl Graz,

Collaborative process maturing support by mining activity streams

iKnow 2015

Use case / context

Background: established research streams

Process mining – the SCHub way

Live demonstration of parts of the solution

Conclusion and outlook

1 2 3 4 5

Agenda

Project: Social Collaboration Hub (SCHub)

Goal: Establishing an integrated infrastructure for

effective support of team collaboration, esp. for

knowledge intensive tasks and regionally distributed employees

• direct support for knowledge and business processes • From a user‘s perspective, a unified intranet

with continuous support for working tasks

without breaches in the workflow should arise.

Solution: Integration of Open Source Software from the areas portal, document management (DMS), groupware and

SCHub System Architecture

Middleware / Supporting Services

CAS nginx Backend Open LDAP MySQL Galera / XtraDB Cluster Postfix Dovecot CEPH End-user Applications

Nuxeo OX App Suite

Liferay Infrastructure Docker neo4j ElasticSearch Camunda BPM Shindig

Knowledge and process maturing

6

Based on Maier, R., Schmidt, A (2007):

Characterizing knowledge maturing: A conceptual process model for integrating e-learning and knowledge management. 4th Conference Professional Knowledge Management-Experiences and Visions (WM’07) , Potsdam.

Goal for SCHub process maturing support

• Allows for collaboratively enhancing document-centric processes

• Use Web 2.0 feature like commenting, rating, tagging in BPM

• Use information available in activity streams to learn about workflows • Support weakly structured processes with software

• Use open standards and available open source software

Focus of the solution

8

Schmidt, A., Hinkelmann, K., Ley, T., Lindstaedt, S., Maier, R., Riss, U.:

Conceptual foundations for a service-oriented knowledge and learning architecture: Supporting content, process and ontology maturing.

In: Networked Knowledge-Networked Media. pp. 79–94. Springer (2009).

communication artefacts

Background: Adaptive Case Management (ACM)

• ACM does not force strict workflows

• Suggests situation-specific actions that might be required • Users can adopt suggested tasks, but can also adapt them • No strict separation between design-time and runtime

• Complements Business Process Management (BPM) • Object Management Group (OMG) Standards:

– Case Management Model and Notation (CMMN, May 2014)

– Business Process Model and Notation (BPMN v.2.0, January 2011) • Camunda supports CMMN since version 7.2 (November 2014)

Example: Oracle Adaptive Case Management

Activity flow for selected task

Background: BPM research streams

• Subject-oriented BPM

– Communication-oriented – Very easy for end-users

– Maybe too simple for knowledge workers • Social BPM

– Bring Web 2.0 participative approaches to BPM

– Potential benefits: increased transparency and knowledge sharing – Drawbacks: possibly lower quality process models, difficult to

Example: Signavio Collaboration Portal

12

Background: process mining and maturing

•

Process Mining

– Discovery type: event logs to process models

– Conformance checking type: compare process model with log – Enhancement type: extend existing models with data from logs

•

Process Maturing

– Weakly structured process might result from missing knowledge

– recording of the users’ activities => creation & maturing of processes – IT support for weakly structured processes is key to link

Example: KISSmir research prototype (Witschel et al. 2010)

Witschel, H.F., Hu, B., Riss, U.V., Thönssen, B., Brun, R., Martin, A., Hinkelmann, K.: A collaborative approach to maturing process-related knowledge.

Own approach

• User activities from all systems are collected in Apache Shindig (OpenSocial) and stored in neo4j (graph database)

• Document creation and publishing timestamp as a frame for the context • Document type, title and headlines are used as semantic context

• Initial case creation

– Suggest tasks and milestones • Case enhancements

Task recommendation algorithm explained

16 document created document v 1.0 published document edited document annotated document reviewed meta data edited

t

t

₀

_t

₁ user group activitites per user pages aggregated task candidates

simi larity

referenced content per activity

Content similarity

• One German and one English corpus for testing

• Two news sites each: heise.de, golem.de, cnn.com, bbc.co.uk • Document consists of selected articles from one source

• Text files with news contents from both sources of the language as reference • Text similarity is calculated for the whole document and for single chapters

• Term frequency * inverse document frequency • For each content, calculate the vector

of all tf*idf values

Live demonstration

Abstracting activities into task descriptions

• Result

– Activities with content that is associated to a chapter of the document – Activity consists of verb, object (content), target (person or system) – List of people that contributed to the document (in-)directly

• If enough activities of a type are found, they have to be aggregated – Results are generic task descriptions

+ content-specific parts based on headlines of the document or titles / keywords of the referenced content

From people to roles

• Possible relations on organizational level – Same department, same level

– Same department, Superior/subordinate – Specific department (e.g. marketing)

– Specific role / job description (e.g., project manager) • Possible relations on the skill level

– Same skill – Different skill – Specific skill set

People similarity / role abstraction

• Information from LDAP user directory are synchronized to neo4j

– Result: graph with roles/job descriptions and org hierarchy derived from „manager“ relationships and department field

• Information from skill management system (Apache Shindig / neo4j) – Result: people – skill tag relationships

• Typical graph operations like shortest path between two people show relation in the organization hierarchy.

Components and their roles

Open LDAP Nuxeo Liferay neo4j ElasticSearch

Camunda BPM Shindig + Mining

BPMN.io

Dovecot

sync org

indexing find relevant emails

report doc events

create case _find

activities visualize & edit cases associate case with doctype

Conclusion

• Combination of manual maturing steps (Web 2.0)

and semi-automatic maturing (activity mining) seems promising

• Major challenges

– Abstraction of concrete activities to general tasks

– Abstraction of concrete users to formal roles

• (German) data protection laws have to be considered

– Our solution does not present person-related data

• Evaluation with real life data is hard to perform

Hof University Alfons-Goppel-Platz 1 95028 Hof, Germany [email protected] www.hof-university.de Phone +49 9281 409-3000 Fax +49 9281 409-4000

BPM + ACM + Web 2.0 + Data Analytics!

Prof. Dr. René Peinl

Head of research group systems integration Teaching area: Web architecture